How do I write a prompt that gets a usable answer?

Prompting is not a trick. It is the on-ramp to everything else, and it is the one skill that travels with you across platforms, tools, and whatever replaces them.

The framework we teach has seven elements. Task: what you want done, stated specifically. Role: who the tool should be while doing it. Context: the situation, the audience, the constraints. Background: the material it needs to work from. Exclusions: what to leave out, which is more powerful than most people expect. Evaluate: read the output against what you actually needed rather than against how impressive it sounds. Iterate: refine and run it again.

Most people do the first element and stop. That is why their results feel random.

The technique that produces the largest single improvement is telling the tool to ask you questions before it answers. Instruct it to keep asking until it is roughly 95 percent confident it understands what you want. You will answer twenty questions. What comes back afterward will not resemble what you were getting before.

Almost every disappointing result starts the same way. Compare these companies. Write a report. Analyze this. The tool then has to guess what you consider important, and it guesses wrong in a fluent and convincing way. What is worth knowing is that this skill transfers. The specific techniques will keep changing, but thinking clearly about what you need before you ask for it is durable, and it works on whatever platform your company standardizes on next.

One more thing worth understanding. Because these systems predict rather than retrieve, the same question can produce a different answer twice. That is not a malfunction. It is a reason to constrain the request rather than to distrust the tool.